Configure the long-term memory plugin for OpenClaw

Comparison

The following two conversations show the difference in behavior between the default Agent and an Agent with the long-term memory plugin enabled in the same scenario.

How it works

The memory plugin runs inside the OpenClaw Gateway. It interacts with the Alibaba Cloud Model Studio long-term memory API through two lifecycle hooks: before_agent_start and agent_end. All read and write operations are sent as HTTPS requests to the Alibaba Cloud Model Studio server-side. Model Studio then handles extraction, vectorization, and semantic retrieval.

• Automatic memory capture (autoCapture): Automatically extracts and stores key information after a conversation ends.

• Automatic memory recall (autoRecall): Automatically retrieves relevant memories and injects them into the context before a conversation begins.

Install and configure the plugin

openclaw gateway status

openclaw plugins install @modelstudio/modelstudio-memory-for-openclaw

{
  "plugins": {
    "slots": {
      "memory": "modelstudio-memory-for-openclaw"
    },
    "entries": {
      "modelstudio-memory-for-openclaw": {
        "enabled": true,
        "config": {
          "apiKey": "sk-xxx",
          "userId": "user_001",
          "profileSchema": "your_profile_schema_id",
          "memoryLibraryId": "your_memory_library_id",
          "projectId": "your_project_id"
        }
      }
    }
  }
}

# View plugin information
openclaw plugins info modelstudio-memory-for-openclaw

# View status
openclaw modelstudio-memory stats

openclaw gateway restart

openclaw plugins info modelstudio-memory-for-openclaw

Memory (Bailian)
id: modelstudio-memory-for-openclaw
Status: loaded
Tools: memory_search, memory_store, memory_list, memory_forget
CLI commands: modelstudio-memory

openclaw modelstudio-memory stats

User: user_001
Total memories: 0
Auto-capture: true
Auto-recall: true
Top-K: 5

The long-term memory plugin is now configured. The Agent's automatic capture and recall features are enabled by default and do not require manual operations. You can open the OpenClaw panel at the default address http://127.0.0.1:18789 to start a conversation with the memory-enabled Agent.

Use memory tools in conversations

In addition to automatic capture and recall, the plugin registers four tools with the Agent. The Agent can actively invoke these tools during a conversation based on the context.

• memory_search: Semantically retrieves memories from the memory library. This tool accepts a natural language query, performs semantic retrieval, and returns a list of the most similar memories. The Agent automatically selects this tool when a user asks retrospective questions, such as "What did we discuss before?" or "What do you remember about the database?"

• memory_store: Writes specified content directly to the memory library without conversational extraction. This is useful when a user explicitly asks the Agent to remember a specific fact, such as "Remember my server IP is 192.168.1.xxx."

• memory_list: Lists all memory entries for the current userId with paging. This tool is useful for browsing and managing existing memories.

• memory_forget: Deletes a specific memory based on its ID. The Agent typically first uses memory_search to locate the target memory and then calls memory_forget to delete it.

Equivalent CLI commands:

# Semantically retrieve memories
openclaw modelstudio-memory search "user preferences"

# List all memories with paging
openclaw modelstudio-memory list --page 1 --size 10

# View memory statistics
openclaw modelstudio-memory stats

Quotas and limits

The Alibaba Cloud Model Studio long-term memory API has the following rate limits:

Performance metrics:

• SearchMemory end-to-end latency: 200–500 ms

• AddMemory latency: 500–1,000 ms

• Automatic capture is asynchronous and does not affect response speed.

FAQ

1. Why is the plugin status not loaded after restarting the Gateway?

   Verify that plugins.entries.modelstudio-memory-for-openclaw.enabled is set to true in openclaw.json. Also, verify that plugins.slots.memory points to "modelstudio-memory-for-openclaw". After you make the corrections, run openclaw gateway restart again.

2. Why do I see an InvalidApiKey error in the logs?

   This error indicates that the DashScope API key is invalid or has expired. Log on to the Alibaba Cloud Model Studio console to verify the API key status and create a new one if necessary. If you are using an environment variable, verify that DASHSCOPE_API_KEY is set correctly and that the Gateway process can read the variable.

3. Can I configure an API key from an Alibaba Cloud Model Studio Coding Plan?

   This is not supported.

4. How do I view the plugin's operational logs?

   The OpenClaw Gateway log files are stored by date in the system's temporary directory. The filename format is openclaw-YYYY-MM-DD.log.

   # Linux / macOS
   tail -f /tmp/openclaw/openclaw-$(date +%F).log | grep modelstudio-memory
   # Windows PowerShell
   Get-Content "$env:TEMP\openclaw\openclaw-$(Get-Date -Format 'yyyy-MM-dd').log" -Wait | Select-String "modelstudio-memory"

   The plugin's automatic capture and recall mechanisms cover most scenarios and do not require extra intervention. You can monitor the growth of memory data in the Memory Library and adjust the topK and cache policy as your business needs grow.

References